The Effect of Transaction Pricing on the Adoption of Electronic Payments

The Eﬀect of Transaction Pricing on the Adoption of Electronic Payments: A Cross-Country Comparison.∗

Wilko Bolt†,DavidHumphrey‡, Roland Uittenbogaard§

October 2005

Abstract Pricing should speed up the substitution of low cost electronic payments for expensive paper-based transactions and cash. But by how much? Norway has explicitly priced individual payment transactions and rapidly shifted to electronic payments while the Netherlands has experienced the same shift without direct pricing. Controlling for differences between countries, we estimate the incremental eﬀect of pricing on the shift to electronic payments. If users strongly value the improved convenience or security of electronic payments, pricing —viewed negatively by most consumers— may not be necessary to ensure rapid adoption of electronic payments. (92 words)

Keywords: electronic payments, transaction pricing, demand elasticity, social beneﬁts

JEL Code: D12, G21

∗Theviewsexpressedinthisarticlearethoseoftheauthors alone and do not necessarily represent those of De Nederlandsche Bank, or the European System of Central Banks. †Wilko Bolt is senior economist in the Research Division at the De Nederlandsche Bank, Amsterdam, The Netherlands, email: [email protected] ‡David Humphrey is professor of ﬁnance at Florida State University, Tallahassee, FL, U.S.A., email: [email protected] §Roland Uittenbogaard is economist in the Payments Policy Division at De Nederlandsche Bank, Am- sterdam, The Netherlands, and is currently employed by the European Commission in Brussels, email: [email protected] 1 Introduction.

The production of electronic payments by banks typically cost from one-third to one-half as much as its paper-based equivalent or cash (c.f., Humphrey, Willesson, Bergendahl, and Lindblom, 2004). As well, merchants’ cost of accepting electronic payments over giro networks and at the point of sale are also lower (credit cards excepted). Since the resource cost of a country’s payments system may account for 2% to 3% of its GDP, it is clear that shifting from paper to electronic payments can confer social benefits. Importantly, the discounted value of these benefits will be larger the more rapidly this shift occurs. There is overwhelming evidence that consumers respond to price incentives but almost no evidence of what this response may be in the payments area. Although consumers are used to responding to price incentives, they tend not to welcome the opportunity to trade off perceived payment preferences with relative prices when their payment use has commonly been viewed as being “free”. While businesses typically pay directly for the payment services they use via explicit transaction fees or compensating balances, consumers have traditionally paid implicitly through lost float or lower (or no) interest on transaction balances. In addition, due to competitive reasons, banks fear a loss of deposit market share if they move first (and are perhaps the only one) to explicitly price consumer payment transactions while anti-trust authorities would be suspicious of industry efforts to coordinate such pricing. One country —Norway— has overcome these difficulties by coordinating only the timing of when direct pricing of consumer payments would start —not the level of prices to be charged which could in fact be zero. The quid pro quo was an elimination of banks’ practice of recouping payment costs through payment float —debiting consumer accounts prior to a value date for bill payments or delaying funds availability for credits to accounts — which made it appear that payment use was “free”. The goal was to make payment costs more explicit so consumers could match better the benefits and costs of different payment instruments, a response expected to lower the social cost of their payment system.1 Our main purpose is to determine the effect of differential transaction-based pricing of payment instruments on the adoption rate of electronic payments. This is done by compar- ing the shift to electronic payments in two countries —one that has pricing (Norway) and one that does not (Netherlands). Transaction-based prices are key since they affect consumers’ decisions about payment use whereas fixed fees are sunk costs ex post that do not vary with usage and thus have limited behavioral effects. The implied discounted social benefitthat follows from a possibly more rapid substitution of low cost for high cost payment instruments is also estimated. Data on payment instrument use for many developed countries is available annually in various Bank for International Settlements and European Central Bank documents, as well as from payment statistics by national central banks. As these time series rarely exceed 15 years, a parsimonious model specification is necessary. A comparable time series of actual payment instrument prices on a broad range of payment instruments is available only for Norway. We contrast the rapid adoption of electronic payments in Norway over 1990-2004 with the experience of the Netherlands which also rapidly adopted electronic payments but did not impose per transaction prices on consumers. By applying a system estimation to our model we are able to improve on the degrees of freedom and increase the

1 Pricing in Norway was implemented by banks individually and encouraged by the central bank. The fact that the larger banks were first to introduce explicit pricing made it easier for the other banks to follow after a lag. Bank efforts to improve the payment system have also occurred in Canada (to eliminate the float incentive to use checks), Germany (to truncate checks and collect them electronically), and the Netherlands (shifting from paper-based credit transfers to “straight-through-processed” direct debits).

2 efficiency of our estimators. If the incremental effect of direct pricing is large, holding constant other within and cross- country influences affecting the adoption of electronic payments, then the potential social benefit can also be large. This would suggest that antitrust concerns raised by possible bank coordination of the implementation of pricing (but not any coordination of the level of prices being charged) could be offset by subsequent social benefits. It would also suggest, in a revealed preference context, that consumers generally place a relatively low value on the greater convenience and security offered by electronic payments since, otherwise, pricing would not have a large separate effect on adoption rates. A final consideration, but one not discussed here, is the loss of seigniorage revenues to the government to the degree that electronic payments replace cash at the point of sale. In what follows, we show in Section 2 how the composition of payments has evolved in Norway and the Netherlands along with the levels of relative prices in Norway. Transaction prices for consumers are zero in the Netherlands.2 Our focus is on the substitution of debit cards for cash (or cash and checks) at the point of sale along with the substitution of (remote) electronic giro payments for paper-initiated giro transactions.3 In Section 3, we specify a parsimonious point of sale “country difference” model to separate the effect of pricing debit card use and ATM cash withdrawals from differences in terminal availability and real personal consumption in our two countries. A similar model relies primarily on prices for the substitution of electronic versus paper-initiated giro payments (since terminal availability is not a constraint). By using two countries we seek to effectively “hold constant” non-price attributes that can influence payment use in addition to pricing when the analysis is applied to only a single country. Our set of four equations are estimated in Section 4 in a seemingly unrelated regression framework to improve efficiency and the effect of prices on payment composition, including the implied price elasticities, are presented. Different models are estimated to judge the robustness of the price effect under alternative specifications, such as different lagged relationships, first differences, and error correction. The social benefitof pricing is illustrated in Section 5 using bank cost data and fitted logistic S-curves to payment use data for Norway and the Netherlands. A summary of our results and the public policy issues they raise are discussed in Section 6.

2 Payment Composition, Pricing, and Other Inﬂuences on Payment Instrument Use.

2.1 Payment Composition. Both Norway and the Netherlands experienced a relatively rapid change in their payment composition for point of sale and bill payment transactions over 1990-2004. Point of sale instruments are now almost solely debit cards and cash but in the early 1990s checks were also important. As seen in Table 1, the number of debit card transactions per person per year in Norway rose from 5 to 146 over our 15 year period, growing by 25% per year.4 The

2 Some fixed fees do exist in the Netherlands. Debit card users pay a flat annual average fee of 6 euro and internet banking has a one-time startup fee of around 15 euro. 3 While check and credit card transactions are included in the analysis, check transactions are only significant in the early 1990s while there are few credit card payments in either country over the whole period. 4 Oil company terminals and cards were introduced in the 1980s as a substitute for cash at gas stations. Although these terminals also accepted bank debit cards, oil company cards could not be used elsewhere and were not priced. The Norwegian payment statistics do not include oil company transactions as debit card

3 Table 1: Payment Instrument Use Per Person in Norway and the Netherlands (1990-2004) 1990 1993 1996 1999 2002 2004 Growth Rate Debit Card transactions Norway 5 16 36 71 113 146 25% Netherlands1 42444667733%

ATM Cash Withdrawals Norway 14 17 22 24 23 22 3% Netherlands 8 20 26 28 30 28 9%

Electronic Giro (credit transfers + direct debits) Norway 15 18 29 46 65 78 12% Netherlands 44 54 70 93 116 124 7%

Paper Giro (credit transfers) Norway 53 44 52 38 24 18 -7% Netherlands343233272118-5% Source: www.dnb.nl, DNB statistics, www.norgesbank.nl, Norges Bank Annual Report on Pay- ment Systems.

Netherlands started from a smaller base of 1 transaction per person per year but rose to 77, a 33% annual growth.5 Part of this difference is due to the fact that Norway started 1990 with far more debit card terminals in place than the Netherlands and so was at a higher point of usage on their logistic growth curve. No time-series data exists on the number of cash transactions, although a few (markedly) different estimates exist for some countries at different points in time. These estimates differ primarily because of the difficulty of estimating very small value cash transactions in which coins are often used and for which stored value cards —which are just starting to gain some acceptance— are the only real substitute.6 We use the number of cash withdrawals at ATMs as our indicator of cash use in transactions. While each cash withdrawal (€138 on average in Norway and €107 in the Netherlands in 2004) funds multiple actual cash transactions, the actofwithdrawingcashispricedinNorwaywhileitsuseatthepointofsaleisnot. Thuswe compare debit card and cash use at the point where both are actually priced and consumer choice is exercised. In both Norway and the Netherlands, debit card use expanded at a rapid rate while growth of ATM cash withdrawals was much smaller. As shown below, the average price of purchases (Norges Bank, 2001, p. 33) and neither do we. Oil company terminals are included in our series of debit card terminals, however, since they accept debit cards for payment. 5 Checks written per person in Norway went from 12 per person annually in 1990 to less than 1 in 2004. In the Netherlands they went from 17 to zero. Credit card transactions per person in both countries were less than 1 in 1990 and only 3 per person (Netherlands) to 5 (Norway) in 2004 (or about 3% of card use in each country). The dominance of debit cards over credit cards is probably due to the fact that banks — not the credit card companies — through a joint venture were the first to introduce EFTPOS directly from deposit accounts and have POS terminals connected to the bank network installed in shops. The banks’ purpose was to replace checks with electronic cards at the point of sale but this also permitted a substitution away from cash. 6 Brits and Winder (2005) provide an estimate of cash use in the Netherlands in 2002. Cash accounted for 85% of POS transactions and 56% of sales while debit cards comprised 13% of transactions and 40% of sales. The average value of a cash transaction was 9.37 euro but over 44 euro for debit cards.

4 an ATM withdrawal rose relative to a debit card transaction in Norway but these two prices were both zero in the Netherlands. If relative price was the only influence on relative use, we would expect a slower growth for ATM withdrawals in Norway (where the ATM price was, after 1996, higher than debit cards) than in the Netherlands (where there is no difference in relative prices). We see indications of thisforATMsinTable1(asgrowthinNorwayis slower than in the Netherlands) but we do not see it for debit card use (where the reverse holds). In order to reflect the substantially lower cost associated with electronic bill payments, employee disbursements, and interbusiness transactions over giro networks, the price of an electronic giro payment in Norway was less than a paper-initiated giro transaction (either delivered in the mail or passed over the counter at a bank or postal office). Since giro prices were zero in the Netherlands, one would expect to see a more rapid growth of electronic giro transactions and slower growth (or greater reduction) of paper giro transactions in Norway than in the Netherlands. Both of these expectations are realized in Table 1. Per person use of electronic giro payments in Norway rose from 15 to 78 over 1990-2004 (growing by 12% a year) while only rising from 44 to 124 over the same period in the Netherlands (growing by 7% annually).7 At the same time, paper giro use fell in both countries but from a higher level and at a greater rate in Norway. Indeed, by 2004 individuals in both countries initiated only 18 paper giro transactions per year.

2.2 Payment Prices in Norway. The average —and sometimes weighted average— per transaction prices being charged for different payment instruments in Norway are illustrated in Table 2. Since there are no per transaction fees in the Netherlands, the relative prices that Norwegian consumers face also reflect the difference in prices faced between Norway and the Netherlands. This is the price effect we wish to separate from other influences on payment choice in these two countries.8 The weighted average per transaction price of a cash withdrawal in Norway was, until 1996, less than that for debit cards. This was because a cash withdrawal at one’s own bank was free during business hours and prices only applied to withdrawals after business hours or at another bank’s ATM. While debit cards started out in 1990 with a price that was more than 3 times higher than the weighted average of different ATM prices (Row 3, Table 2), it ended up being only 65% of the cash withdrawal price in 2004. Thus only after 1996 did the absolute price of a debit card favor its use over cash when EFTPOS terminals were available. But even before 1996, there was an indirect inducement to use debit cards in Norway when it became possible in late 1992 to obtain “cash-back” from a debit card transaction at the point of sale. This avoided the extra cost and inconvenience of having to use an ATM to withdraw cash since small amounts of cash could be obtained at no additional cost when making purchases at the local market. There was a stronger relative price inducement to use an electronic rather than a paper- initiated giro transaction for consumer bill payments. In 1990, the price of an electronic giro transaction was only 29% as high as a paper giro payment but by 2004 this had fallen to only 9% of the paper price. In the beginning, electronic giro payments were initiated via

7 By 2004, direct debits accounted for 10% of electronic giro payments in Norway but 56% in the Netherlands. This is the main reason why electronic giro payments per person in the Netherlands are so much higher than in Norway. 8 Consumers in the Netherlands do face a ﬁxed annual fee for the use of a debit card but, as the fee is ﬁxed, the consumer ”sees” a zero price per additional transaction and responds accordingly.

5 Table 2: Average Per Transaction Prices for Diﬀerent Payment Instruments in Norway Prices in Euros 1990 1993 1996 1999 2002 2004 Growth Rate Debit Card Price Norway .18 .23 .25 .26 .28 .26 2%

ATM Cash Withdrawal Price Norway .05 .18 .24 .29 .39 .40 14%

Relative Price: Debit Card/ATM Cash Withdrawal 3.60 1.28 1.04 .90 .72 .65 -11%

Electronic Giro Price Norway .10 .18 .22 .23 .31 .27 7%

Paper Giro Price Norway .35 .62 1.18 1.86 2.65 2.76 15%

Relative Price: Electronic Giro/Paper Giro .29 .29 .20 .12 .11 .09 -8% Source: www.dnb.nl, DNB statistics, www.norgesbank.nl, Norges Bank Annual Report on Pay- . ment Systems. telephone but this was later overtaken by the spread of internet banking. This applies to credit transfers where the consumer retains control in initiating a payment, as opposed to a direct debit where the receiver of the credit initiates the debit to the consumer’s account under a prearranged contractual agreement. It is important to note that the prices charged in Norway do not cover the full bank cost of making a payment (c.f., Flatraaker and Robinson, 1995; Gresvik and Øwre, 2003). In 1988, transaction prices covered only around 25% of the banks’ payment cost but this coverage had risen to around 70% in 2001.9

2.3 Terminal Availability and Levels of Consumption. While relative prices provide an inducement to use electronic payments at the point of sale, this can only be accomplished if a merchant has an EFTPOS terminal that can be used. This observation points to the two-sided nature of the payment market which influences the adoption rate of new payment instruments. In particular, the market for electronic payment services is considered a two-sided market in the sense that both consumers and merchants are needed simultaneously to demand and “consume” card payments. Suppliers of payment card services (or so-called “platforms”) can effectively cross-subsidize between merchants and consumers through differential pricing to stimulate this demand. In two-sided markets, typically only one side is charged on a transaction basis while the other side obtains the service (almost) for free in order to generate greater demand.10 Indeed, merchants value a wide diffusion of payment cards among consumers while consumers benefit from high terminal

9 The relationship between fees and underlying costs is diﬀerent in Sweden with surplus bank revenues from card transactions cross-subsidizing the expense of providing cash, distorting resource allocation (Sveriges Riksbank, 2004). 10In Norway, the consumer side is directly charged for its use of payment instruments while in the Nether- lands the retailer side of the market pays per transaction. Bolt and Tieman (2004) provide an explanation

6 Table 3: Terminal Availability, Real Consumption, and Demographic Inﬂuences on Payment Instrument Use 1990 1993 1996 1999 2002 2004 Growth Rate Debit Card EFTPOS Terminals (per mil population) Norway 2,487 6,324 8,932 13,214 17,723 21,091 15% Netherlands 148 1,600 6,170 9,176 10,941 11,967 34%

ATM Terminals (per mil population) Norway 419 396 426 451 484 473 0.8% Netherlands 180 291 395 421 465 468 6.6%

Real Per Capita Personal Consumption (in 000) Norway 11.9 12.1 13.9 15.1 17.8 16.7 2.3% Netherlands 9.2 9.3 9.9 10.9 11.4 11.3 1.4%

Share of Young Adults in Population Norway 8.0 7.8 7.2 6.4 6.0 6.0 -1.9% Netherlands 8.5 8.2 7.0 6.1 6.0 6.0 -2.3% Source: www.dnb.nl, DNB statistics,National Accounts, Dutch CBS, www.norgesbank.nl, Norges Bank Annual Report on Payment Systems, IFS. density at retail locations that accept their cards. In our analysis, payment card and ATM terminal density are included to take this two-sided effect into account in explaining relative payment card usage. Table 3 shows the number of EFTPOS terminals in place in Norway and the Netherlands over 1990-2004 per one million of population (which controls for differences in population size).11 Asshowninthefirst two rows, Norway had almost twice as many debit card terminals as the Netherlands in 2004 and this difference was far more extreme in earlier periods. While the growth of EFTPOS terminals has been more than twice as rapid in the Netherlands, it still has a long way to go to provide the same density of terminal access as Norway. By this measure alone, it really would not be possible —regardless of any price incentive— for consumers in the Netherlands to use debit cards with the same intensity per person as they do in Norway. As noted earlier, there is no price incentive to use debit cards in the Netherlands so there are two reasons —no price incentive and fewer EFTPOS terminals per person— to expect that the Netherlands would use debit cards less intensively than in Norway. Even so, as shown below, it is difficult to separate the effect of prices from terminal availability on debit card and ATM use. The same “separation problem” exists for cash withdrawals at ATMs. Norway prices ATM withdrawals while the Netherlands does not and for the entire period Norway also provided a greater density of ATMs to withdraw cash from (Row 3, Table 3). Separating the price effect from the terminal effect for ATM cash withdrawals may be somewhat easier here since by 2004 both countries had almost the same ATM density but withdrawals were priced only in Norway and, compared to the Netherlands after 1993, per person use in Norway was correspondingly less (Row 3, Table 1). Inferences on the relative importance of pricing may be more accurate if two other possible, for these widely observed completely skewed pricing strategies in two-sided markets. See Rochet and Tirole (2003) for a rigorous analysis of two-sided markets and competition. 11In 2004, the population in the Netherlands was 16.3 million while in Norway it was only 4.6 million.

7 but small, influences on payment choice are considered. One concerns differences in the level of real per capita personal consumption between the two countries, since higher levels of real consumption tend to be associated with larger numbers of transactions.12 Asecondinfluence concerns the possibility that changes in the number of young adults in both countries may affect differences in new payment adoption rates. Consumer surveys indicate that young adults and higher income individuals adopt new payment arrangements more rapidly than others even without pricing. But direct pricing could well affect the adoption rates of those with greater habit persistence, a lower opportunity cost, or who do not value much the added convenience or security that electronic payments can offer.13 The level and variation of both per capita consumption and the share of young adults in the population over time are illustrated in the bottom half of Table 3. Real per capita consumption in Norway was 29% greater than that in the Netherlands in 1990 but rose to be 48% higher in 2004. This difference should be associated with a rising number of all types of transactions in Norway relative to the Netherlands. There are smaller differences between these two countries in the shares of young adults —new entrants into the labor force aged 20 to 24. Indeed, these shares are falling in both countries.14

3ACountry-Diﬀerence Model of Payment Choice.

3.1 A Country-Difference Model. Differences between Norway and the Netherlands are used to try to explain per capita use of debit cards, ATM cash withdrawals, and electronic and paper giro payments. As outlined above, the main influences on payment use and composition are differences in the number of EFTPOS and ATM terminals per million population, the prices being charged in Norway (positive) and the Netherlands (zero), and differences in the level of real per capita consumption. Our time period is short (only 15 years) as time-series data on payment instrument use has only recently been deemed important enough to be routinely collected at the country level by government agencies. While some time-series on some payment types do exist for longer periods in some countries, this information is not comprehensive nor are payment instrument prices available since very few types of payment services are directly priced. Norway is the exception that allows us to undertake this analysis. These data constraints impose a parsimonious specification on our explanatory four equation model:

CARDt = α1 + α2CARDT ERMINALt 1 + α3CARDPRICEt − 2 2 +1/2(α22CARDTERMINALt 1 + α33CARDPRICEt ) − (1) + α23CARDTERMINALt 1 CARDPRICEt − ∗ + α4CONSUMPTIONt + ε1t

12All monetary values for Norway (prices as well as real consumption) have been translated from Norwegian kroner into euros using a purchasing power parity exchange rate. Also, real per capita consumption in Norway includes oil revenues only indirectly as some of this revenue is used to ﬁnance government expenditures which likely reduces taxes from what they would otherwise be —permitting real consumption to be larger. 13Zinman (2004) develops a model of implicit costs to explain why debit cards are replacing credit cards in the US. 14In implementation, this variable was deleted since it created convergence problems in our system estima- tions due to being an extremely smooth series.

8 AT Mt = β1 + β2ATMTERMINALt 1 + β3CARDPRICEt − 2 2 +1/2(β22AT MT ERMINALt 1 + β33CARDPRICEt ) − (2) + β23ATMTERMINALt 1 CARDPRICEt − ∗ + β4CONSUMPTIONt + ε2t

2 EGIROt = γ + γ EGIROPRICEt +1/2(γ EGIROPRICE ) 1 2 22 t (3) + γ3CONSUMPTIONt + ε3t

2 PGIROt = δ1 + δ2EGIROPRICEt +1/2(δ22EGIROPRICE ) t (4) + δ3CONSUMPTIONt + ε4t.

In the variable deﬁnitions, NOR indicates Norway and NL indicates Netherlands and diﬀerences between these countries are expressed in index form:15

CARD = ln (NOR debit card use/NL debit card use), on a per person basis; CARDTERMINAL = ln (NOR card terminals/NL card terminals), per million population; CARDPRICE = ln (NOR card price/NOR ATM price); CONSUMPTION = ln (NOR personal consumption/NL personal consumption, real per capita; AT M = ln (NOR ATM cash withdrawals/NL ATM cash withdrawals), per person; AT MT ERMINAL = ln (NOR ATM terminals/NL ATM terminals), per million population; EGIRO = ln (NOR electronic giro use/NL electronic giro use), per person; EGIROPRICE = ln (NOR electronic giro price/NOR paper giro price); PGIRO = ln (NOR paper giro use/NL paper giro use), per person.

As debit card and ATM terminals have to be in place before consumers can use them, and even when in place typically have a lag before they are used at a significant level, these two terminal variables are lagged by one year in the model to give a closer correspondence between the exogenous availability of new terminals and their possible effect on use. Prices, of course, also have to be known before they can affect payment choice. The lag here is likely much shorter and prices are specified as exogenous and contemporaneous.16

15In many cases, the log of the absolute difference in our variables between countries was negative (or changed from positive to negative) so all variables are expressed as the log of the ratio or index of the difference between countries. 16The effects of different lag arrangements on the results are noted in Section 4.

9 3.2 Illustrating the Determinants of Payment Composition. While our model looks different from the usual demand equation specification used to estimate price effects on payment instrument use in a single country (c.f., Humphrey, Kim, and Vale, 2001), the same variables —quantity of use, levels of own price, substitute price, and income (via consumption)— are included in the analysis. Ideally, we would like to run an experiment where there are no price inducements to affectconsumerchoiceofpaymentinstrumentsin a country, observe the rate of adoption of electronic payments based solely on non-price considerations along with terminal availability (if required), and then re-run the experiment adding price inducements in order to separate price from non-price effects. This would address the omitted variable problem of unknown non-price attributes of different payment instruments.17 Our use of two countries —one with pricing, one without— addresses the non- price attribute omitted variable problem if country differences in non-price attributes are relatively small, as seems likely. In any case, we also estimate the effect of pricing in one country (Norway) below but are aware that these results may incorporate unknown non-price attributes. The observed relationships underlying our four equation model are illustrated in the two figures below. In Figure 1, the price of debit cards in Norway is steadily falling relative to the weighted average price actually incurred for an ATM cash withdrawal over 1990-2004 (line with circles). Although the relative price of debit cards was falling, the absolute price was higher than a weighted average of ATM cash withdrawal prices up until 1996 (Table 2). Thus it is not surprising that relative debit card use in Norway compared to the Netherlands was falling (as seen in the top line with boxes) up until 1997.18 After 1996-97, the absolute price of debit cards in Norway was less than an ATM cash withdrawal19 and debit card use in Norway expanded slightly faster than in the Netherlands (which did not have this price inducement). While a falling relative price is expected to promote use, this apparently occurred only when the price of a debit card transaction was absolutely less than an ATM withdrawal. Overall, the pattern of debit card and ATM prices in Norway reduced debit card use relative to the Netherlands up to 1996 and expanded it slightly afterwards (shown as a slight rise in the top line with boxes after 1996). Thus we should not expect a strong price elasticity response for debit card use. And, indeed, the response we obtain is weak. The situation for ATMs is essentially the reverse. The average ATM price is lower than debit cards in Norway over 1990-1996 so ATM use in Norway may be expected to exceed that in the Netherlands where these prices are the same (both at zero). But ATM use in Norway expands at a slower rate than in the Netherlands up until 1995 (as the bottom line with boxes falls over this period). After 1998 when the debit card price is absolutely lower than the ATM price, Norwegian ATM use falls slightly relative to the Netherlands. Consumer choice seems to be affected by absolute prices as well as changes in relative prices. For debit cards as well as ATMs, use seems to closely follow the availability of terminals needed to function these transactions. This implies a positive relationship. As shown below, however, ATM terminal availability appears to have gotten ahead of terminal use in both countries toward the end of our period, leading to a negative relationship where

17For example, if an electronic payment instrument offers significant convenience or security benefits compared to its paper-based alternative, then pricing the paper instrument higher than its electronic alternative would likely generate a larger price response than if such non-price attributes did not exist. 18The curves in all figures are cubic splines of plotted actual observations. 19This is where the ratio of the debit card to ATM price equaled one so the log of this value (the line with circles) was at zero in the figure.

10 Figure 1: Debit Card Use, Prices, and Terminal Availability

3.5 Lines with boxes: Card or ATM use = ln (Nor DCARD or ATM use/Nl DCARD or ATM use) 3.0 Card Terminals Solid line: Number of Card or ATM Terminals = ln (Nor terminals/Nl terminals) 2.5 Line with circles: Relative Card Price = ln (Nor DCARD price/Nor ATM price) 2.0

1.5 Card use

1.0

0.5 ATM Terminals 0.0 ATM use

-0.5 Relative Card Price

-1.0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

terminals expand but use falls. The price-use relationship for electronic and paper giro transactions seems clearer, perhaps because there are no “terminal constraints” and likely weaker non-price attributes to potentially confound the eﬀect of prices on relative giro use.20 In Figure 2, the price of an electronic giro transaction in Norway falls relative to paper only after 1993 (line with circles). Importantly, and unlike the debit card to ATM price relationship, the electronic giro price is always absolutely lower than the paper price in Norway. Thus use of electronic giro payments in Norway is seen to rise relative to that in the Netherlands over the whole period (bottom line with boxes) while the relative use of paper giros falls —although only slightly— after 1995 (top line with boxes). The relationships illustrated in Figure 2 indicate that the relative level of payment prices, as well as the change in relative prices over time, are important in determining relative payment use for giro payments. The same holds for debit card and ATM use in Figure 1.

4 Estimation results and the Eﬀect of Price on Payment In- strument Use.

The system of equations (1) to (4) was estimated in a seemingly unrelated regression framework to allow for the possible correlation between errors in locally identifying debit card use

20While an internet connection is needed for consumers to make an electronic credit transfer, this is not needed for a direct debit. As well, the number of internet connections far exceeds the number of internet banking users so an internet connection does not act as a binding “access constraint” in the same manner as a debit card or ATM terminal. Obviously, the same holds for telephone connections and telephone giros.

11 Figure 2: Electronic and Paper Giro Use and Prices

1.0

Paper Giro use = ln (Nor paper giro use/Nl paper giro use) 0.5

0.0

-0.5

-1.0 Electronic Giro use = ln (Nor electronic giro use/ Nl electronic giro use)

-1.5 Relative Giro Price = ln (Nor electronic giro price/ Nor paper giro price) -2.0

-2.5 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

with that for ATM cash withdrawals and similarly for electronic and paper giro use. The estimated parameters and their associated t-statistics are shown in the Data Appendix. With 15 observations per equation, there are 38 degrees of freedom (d.f. = 4*15 - 22). As expected from viewing Figures 1 and 2, the explanatory power of the model was high (the respective R2s were .98, .99, .87, and .80 from the system estimation). While the variables are not I(0) and only infrequently bivariately cointegrated, the Durbin-Watson statistics suggest no positive autocorrelation of residuals (the D-W values were, respectively, 2.12, 2.80, 1.86, and 2.03). In addition, the relationships shown in the figures appear to be reasonable and expected from theory so the degree of spurious correlation, if any, is likely to be small. Even so, we report results using first differenced data as well as an error correction model (which supports our results using levels data).

4.1 The Effect of Price on Payment Use. The derivatives of equations (1) to (4) with respect first to relative prices and then with respect to lagged terminal availability are shown in Column 1 of Table 4 (and they are all significantly different from zero). A 10% reduction in the price of debit cards relative to an ATM cash withdrawal is associated with a 2.2% rise in the relative use of cards in Norway compared to the Netherlands (which has a zero explicit price for both cards and ATMs). At the same time, at 10% increase in debit card terminals in Norway relative to the Netherlands is associated with a 5.3% rise in debit card use in Norway relative to the Netherlands. As seen in the table, lagging both terminals and prices by one period doubles the strength of the price response (from -.22 to -.48) but does not alter the terminal elasticity. Assuming no lags, however, increases considerably the apparent responsiveness of debit card use to changes

12 Table 4: Price and Terminal Elasticites Under Different Model Specifications Lagged Lagged Terminals No Lags Separate Prices Terminals and Prices Own Substitute Debit Card Price Effect -.22* -.48* -.06 -.19 -.03 Terminal Effect .53* .57* .94* .49*

ATM Cash Withdrawal Price Eﬀect .23* .31* .29* -.85* .69* Terminal Eﬀect -.16* -.49* -.35* -.35*

Electronic Giro Price Eﬀect -.46* -.53* -.44 .21* .10*

Paper Giro Price Effect .27* .25* .33* -.03 .03 Starred (*) values indicate a P-value smaller than .01 in a two-tailed t test. in terminal availability—making it almost one-to-one in percentage terms—but the trade-off is to generate a price elasticity insignificantly different from zero. Over 1990-2004, the price of ATMs in Norway rose relative to debit cards. The price elasticity suggests that a 10% rise in the relative price of ATMs is associated with a 2.3% decrease in relative use.21 Numerically, this is very similar to the result for the debit card equation where a 10% reduction in the relative price of debit cards gives a 2.2% rise in relative use.22 The ATM terminal elasticity, however, has an unexpected sign and is negative at its mean. When evaluated yearly, the terminal elasticity is positive over 1990-1994 but the negative relationship for the remaining years dominates, giving a negative mean. Looking more closely at ATM use and terminal availability (Tables 1 and 3) suggests that the source of the negative elasticity is that per person ATM use in Norway reaches a peak in 1998 and then falls while ATM availability in Norway reaches a peak five years later in 2003. Similarly, ATM use in the Netherlands peaks in 2001 but terminals continue to expand. The apparent explanation for the negative ATM terminal elasticity is that ATM use has reached saturation (due in part to the price disincentive) while terminals are still being added, giving the result that terminals are expanding while use is falling. The estimated price effects for electronic and paper giro payments conform to expectations since, when the relative price of electronic giro transactions falls by 10%, relative use of this instrument in Norway rises by 4.6% compared to the Netherlands. Similarly, a 10% increase in the relative price of paper giro payments is associated with a 2.7% reduction in relative use between the two countries.23 Our preferred model in equations (1) to (4) was respecified so that direct debits, which comprise 10% of electronic giro payments in Norway but 56% in the Netherlands, were deleted from the electronic giro use and price data. This had almost no effect on the results shown in Column 1 of Table 4. Equations (1) to (4) were respecified

21Since the price ratio used in the ATM equation is the same as that used in the debit card equation—ln (Norway debit card price/Norway ATM price)—the negative debit card price elasticity would become a positive elasticity in the ATM equation. 22The ATM price eﬀect is larger when both terminals and prices are lagged in the model. 23As the same price ratio is used in both the electronic and paper giro equations—ln (Norway electronic giro price/Norway paper giro price), the negative electronic giro price elasticity would become a positive elasticity in the paper giro equation.

13 again to include checks with ATMs so that both can substitute with debit cards. Checks were important in the early 1990s, had a high price, and their use effectively fell to zero by 2004. Nothing of substance was changed except that the debit card price elasticity lost significance. Real per capita personal consumption was markedly higher in Norway and growing faster than in the Netherlands. We expected that this would have a significantly positive effect on expanding relative electronic payment use in Norway. However, as seen in the Data Appendix where the parameter results are shown, the effect of real per capita personal consumption on payment use was insignificant in all four equations. Just as an exercise, equations (1) to (4) were simplified by deleting the squared terminal, squared price, and terminal-price interaction variables. Then the remaining price ratio in each equation (e.g., debit card price/ATM price and electronic giro price/paper giro price) was reexpressed as the log of separate own and substitute price variables for each equation. The resulting own and substitute price elasticities, along with the reestimated terminal effect, are shown in the last two columns of Table 4. Our preferred model (in Column 1) is specified in ratio form, due to our limited sample, but it is of interest to see the implied own and cross- price elasticities that result from estimating each price elasticity separately. All but one own price elasticity is negative and three of the four cross-price elasticities are positive (as would be expected for a substitute payment instrument). However, considering that only one negative own elasticity and two positive cross elasticities were significant, it seems that the price effects are not very strong. An earlier study of the effects of payment instrument pricing in Norway over 1989-1995 (Humphrey, Kim, Vale, 2001) found significant and inelastic own price elasticities for debit cards (-.35) and ATM cash withdrawals (-.55) along with significant substitution between debit cards and ATMs (which ranged from -.11 to -.46). Although this study did not control for terminal availability as we do, the results obtained do not greatly differ from ours. In both studies, own price elasticities are all inelastic and debit cards significantly substitute for ATM cash withdrawals with an elasticity less than .50.

4.2 An Error Correction Specification. Although our model is specified in log-differences between Norway and the Netherlands, the variables used in our system estimation are non-stationary. As the estimation results may be biased and could reflect spurious correlations, our model was reestimated in first difference and error-correction forms. The model in levels, in first differences, or in error correction form are all nested within an “autoregressive distributed lag” framework. The imposed restrictions in this framework are linear and easily tested. To illustrate, consider the following extension of equation (1), written in an autoregressive distributed lag regression format:24

CARDt = α + γCARDt 1 + δ1CARDPRICEt + δ2CARDPRICEt 1 (5) − − + β1CARDT ERMINALt 1 + β2CARDT ERMINALt 2 + ut. − −

The simultaneous restrictions γ = β2 = δ2 =0brings us back to a model speciﬁed in (current) levels only. Without aﬀecting its ability to explain the data or changing the least squares

24Note that compared to equation (1), consumption, the squared variables, and interaction terms have been excluded. These additional variables could be included without aﬀecting our illustration.

14 estimates of the parameters of interest, (5) may be rewritten as:

∆CARDt = α +(γ 1)CARDt 1 + β1∆CARDTERMINALt 1 − − − +(β1 + β2)CARDT ERMINALt 2 + δ1∆CARDPRICEt (6) − +(δ1 + δ2)CARDPRICEt 1 + ut. −

Imposing the simultaneous restrictions γ =1, β1 + β2 =0,andδ1 + δ2 =0, generates a model specification in first differences which can be tested using a Wald test. Alternatively, (5) may be rewritten in error correction form:

∆CARDt = α + β1∆CARDTERMINALt 1 + δ1∆CARDPRICEt − +(γ 1)(CARDt 1 λ1CARDT ERMINALt 2 (7) − − − − λ2CARDPRICEt 1)+ut, − −

β1+β2 δ1+δ2 where λ1 = γ 1 and λ2 = γ 1 . − − If the parameter γ1 1 is negative and signiﬁcantly diﬀerent from zero, the model in error correction format cannot− be rejected. In (7) we have an equilibrium relationship:

∆CARDt = α + β1∆CARDTERMINALt 1 + δ1∆CARDPRICEt + ut, − and an equilibrium error,

CARDt 1 λ1CARDT ERMINALt 2 λ2CARDPRICEt 1, − − − − − that measures the extent to which the long-run relationship between the variables CARDt 1, − CARDT ERMINALt 2 and CARDPRICEt 1 is not satisfied. Consequently, the feedback parameter γ 1 can be− interpreted as the proportion− of the resulting disequilibrium that − is reflected in the movement of CARDt in one period. If specified in logarithms, the parameters λ1 and λ2 yield estimates of long-run elasticities which would be equivalent to the price and terminal elasticities we derived from equations (1) to (4) if our model in levels is appropriate. Interestingly, error-correction models are also appropriate when the variables are non-stationary and are particularly attractive when the dependent variable is I(1). Gen- erally, the significance of the parameter γ 1 indicates the existence of a long-run equilibrium and the cointegration of the non-stationary− variables.25 This would allow direct estimation of our preferred model in levels. The error correction results of all four equations are shown in Table 5 along with the price and terminal elasticities for our preferred model from Table 1 using levels data along with new elasticities using only first differenced data. In a first difference framework both terminal elasticities have the expected sign and are significant, but this is at the expense of poor results for the price elasticities. In the error-correction form, the debit card price elasticity is no longer significant but the other price elasticities have the expected sign and are significant (even with a reduction in degrees of freedom), although in two cases the feedback parameter was not significant at the 5 percent level. Given our data limitations, the results weakly suggest that the price elasticities using levels data in equations (1) to (4) —our preferred model— are robust and can be relied upon as long-run estimates.

15 Table 5: Price and Terminal Elasticites: Data in Levels and First Differences Levels 1st Differenced 1st Differenced & Levels Data Data Data in an Error Correction Model Debit Card Price Effect -.22* .47 3.25 Terminal Effect .53* .75* .82 Feedback Parameter -.066

ATM Cash Withdrawal Price Eﬀect .23* .06 .32* Terminal Eﬀect -.16* .47* -.39* Feedback Parameter -.826*

Electronic Giro Price Eﬀect -.46* .37* -.60* Feedback Parameter -.240

Paper Giro Price Effect .27* .11 .49* Feedback Parameter -.448* Starred (*) values indicate a P-value smaller than .01 in a two-tailed t test. Debit cards and ATMs formed one system estimation while electronic and paper giros formed another in the first differenced and error correction models.

4.3 Estimation of Electronic for Paper Substitution in Norway. The effect of pricing on payment instrument use is also estimated for Norway alone. This approach should give similar results to our country difference model if non-price characteris- tics that affect payment use in a country are not too strong. The specification is linear and simpler than our country difference model (due to degrees of freedom considerations) and all the data are for Norway:

CARDATMt = α1 + α2CARDATMTERMINALt 1 (8) − + α3CARDAT MP RICEt + α4CONSUMPTIONt + ε1t

ELEPAPERt = β1 + β2ELEPAPERPRICEt + β3CONSUMPTIONt + 2t (9)

where:

25As a stability condition, the feedback parameter γ 1 needstobebetweenzeroand-1. −

16 Table 6: Price and Terminal Substitution for Norway Debit Card/ATM substitution: Levels Data 1st Differenced Data Price Effect -.20* -.31* Terminal Effect .54* .06

Electronic/Paper Giro Substitution: Price Eﬀect .54* .13 Starred (*) values indicate a P-value smaller than .01 in a two-tailed t test.

CARDATM = ln (debit card use/ATM use), on a per person basis; CARDATMTERMINAL = ln (card terminals/ATM terminals), per million population; CARDAT MP RICE = ln (debit card price/ATM price); CONSUMPTION = ln (personal consumption), real per capita; ELEPAPER = ln (electronic giro use/paper giro use), per person; ELEPAPERPRICE = ln (electronic giro price/paper giro price).

Equations (8) and (9) were estimated in a systems equation framework (with 23 degrees of freedom, d.f. = 2*15 - 7). As shown in Table 6, the elasticity of substitution between debit cards and cash was -.20 using levels data and -.30 in first differences. A 10% rise in the relative price of an ATM cash withdrawal (which reduces the ratio of debit card to ATM prices) is associated with a small (2.0% or 3.1%) rise in the ratio of debit card to ATM use. If this parameter was -1.0, then the expenditure shares of debit cards and ATMs would be unchanged as a 10% relative rise in the ATM price would be exactly offset by a 10% decrease in relative ATM use. Since the parameter is less than one (in absolute value), the expenditure share of ATMs rises as the price induced substitution is less responsive than in (say) a traditional Cobb-Douglas framework where the elasticity of input substitution to a price change is 1.0. The elasticity of terminal availability on debit card and ATM use is .54 which indicates that a 10% relative rise in debit card terminals leads to a 5.4% relative rise in debit card use. The substitution elasticity between electronic and paper giro transactions had the wrong sign (at .54) and was significant using levels data in a system estimation. Use of first differenced data did not alter this sign but dropping real personal consumption from (8) and (9), which had a large and significant effect, did (giving a significant -1.98 value for electronic/paper giro substitution).

4.4 Conclusions Regarding the Effect of Pricing on Payment Use. Three conclusions can be drawn from our price elasticity results so far. First, in Table 4 and in our alternative specification here, the effect of terminal availability on relative debit card and ATM use exceeds that for pricing since the terminal elasticities are larger. This implies that convenience, safety, and other non-price attributes of different payment instruments are themselves an important inducement to change payment use, as long as terminals are available, than is price. Even so, pricing does have a significant effect in

17 influencing payment choice but not as much as we had expected. A second conclusion, which follows in part from the first, is that changes in relative prices (or terminals) both have a smaller than proportional effectonrelativeuse. Inthissense,botheffects are relatively inelastic. Third, the similarity of the price elasticity results from our country difference model with those for debit card/ATM substitution in Norway alone, suggest that the omitted variable of unspecified non-price payment instrument attributes is not strong enough to markedly distort price elasticity results derived from a single country estimation. Although approximate, our overall conclusion is that while terminal availability appears to have a stronger effect on relative payment instrument use than does direct per transaction pricing, the shift to electronic payments could be speeded up when pricing is combined with terminal availability. If both prices and terminals are expanded at similar percentage rates then the adoption of electronic payments could have been speeded up by perhaps 40% compared to not having per transaction pricing.26 As seen in Tables 2 and 3, however, debit card terminals changed at a much greater rate than did the price of ATMs or the relative prices of cards to ATMs, indicating that in this instance a potential speedup of 40% is too high and was not realized. More precisely, for Norway the average annual growth in debit card terminal density equaled +15%, whereas the growth in card price relative to ATMs was -11%. Given the estimated elasticities in Table 4, this would predict a relative rise of debit card use over ATMs of 15 0.53% + 11 0.22% = 10.4% from the terminal and price effects alone. Without any price× inducements× this increase in usage would be 15 0.53% = 8.0%, suggesting that the substitution process has been speeded up by approximately× 2.4/10.4 = 23%, although the realized contribution of pricing to debit card adoption was 2.4% a year.27 Electronic giro payments do not have a terminal constraint and the influence of consumption growth on payment use is not significant so only the effect of pricing is measured. The growth of electronic giro relative to paper giro prices was -8% while the price elasticity in Table 4 was .46, suggesting that the realized contribution of pricing to the adoption of electronic giro payments was 8 .46 = 3.7% annually. Thus in terms of both the size of the estimated price elasticities× and their realized impact on adoption rates, the effect of pricing on the shift to electronic payments is greater for giro transactions than for debit cards. The fact that terminal elasticities are an important component of the substitution process for cards suggests that non-priced attributes —such as convenience and security —play a greater role for cards versus cash than for electronic versus paper giro transactions.

5 Illustrating the Social BeneﬁtofElectronicPayments.

Logistic or “S-curves” have been successfully used to approximate the adoption and dispersion of new products and technology, such as the adoption of the telephone, the television, the internet, and the use of robots in manufacturing. This procedure is applied here to electronic payments since, during our 15 year period, the shares of debit cards (in total debit card and ATM transactions) and electronic giros (in electronic plus paper giro payments) made the transition from rising at an increasing rate to rising at a decreasing rate (c.f., Table 7).28

26This estimate is derived from the ratio of the price elasticity in our preferred model (-22%) in Column 1 of Table 4 to the terminal elasticity (53%), which equals .42. 27The same calculation using price (-.20) and terminal (.54) elasticities for Norway alone from Table 6 gives a 20% speedup for debit card use (with a contribution of 2.2% annually). 28The coverage of an inﬂection point for electronic giros in the Netherlands is weak, however.

18 Table 7: Debit Card and Electronic Giro Payment Shares in Norway and the Netherlands Debit Card Electronic Giro Norway Netherlands Norway Netherlands 1990 .27 .12 .22 .56 1995 .58 .39 .34 .67 2000 .78 .63 .60 .80 2004 .87 .73 .82 .87 Percentage point change in share per year: 4 4.1 4 2.1 Payment share are: debit card use/(debit card + ATM use) and electronic giro use/(electronic + paper giro use).

Over 1990-2004, the average change in debit card shares at 4 percentage points per year in both countries suggests that per transaction pricing of debit card and ATM transactions in Norway did not appear to speed up the adoption of cards. If it had, we would expect to see a larger average share change for Norway than for the Netherlands (rather than the reverse seen in Table 7). This result is essentially consistent with the low estimated price elasticities for debit cards compared to electronic giros seen in Tables 4 and 5.29 The smaller change in electronic giro shares for the Netherlands compared to Norway implies that pricing giro transactions has speeded up the transition to electronics. For giro transactions, the relative price of paper giros was rising but, in addition, the electronic price was absolutely lower than the paper price. Although the relative price of ATM transactions was also rising over our sample period, the debit card price was absolutely higher than the ATM price for the beginning one-third of our sample. While economists maintain that only changes in relative prices matter, common sense suggests that consumers are also influenced by absolute prices. Our data shows a falling relative price of debit cards over the whole period but, if the debit card price had remained above the weighted average of an ATM transactionfortheentireperiodaswell,itisveryunlikelythatmanyconsumerswouldhave chosen to use a debit card unless the non-price attributes were very strong. Since inflection points are covered in the data series, logistic S-curves were estimated and are shown in Figures 3 and 4. The S-curves are from: ln(St/(1 St)=a + bT + t where − St = the share of electronic payments in total electronic and paper transactions, T = time, and b is the slope of the S-curve. This form imposes symmetry around the inflection point in the data and assumes that electronic payments will, at some point in time, completely replace their paper-based alternative. This is more likely for electronic versus paper giro transactions than it is for debit cards and cash, unless stored value cards become popular and replace cash use.30 The fit to the observed data can be assessed by the extent to which the data points over 1990-2004 deviate from the solid curves in the figures.31 All the other

29Indeed, replacing the dependent variables in equations (1) to (4) with the index of shares, rather than the index of per person use, reverses the sign for the debit card price elasticity and gives an insignificant paper giro price effect. With the index of shares—ln [(Norway share electronic/share paper)/(Netherlands share electronic/share paper)] the price effects are weaker than those we identified earlier. 30 A non-linear, symmetric, logistic curve specification exists that can impose an upper limit (say St∗ = .90)to the replacement of ATM cash withdrawals by debit cards using ln St = St∗/(1+c exp( bT ))+t. Unfortunately, we were unable to obtain convergence when this model was estimated. − 31The adjusted R2s ranged from .91 to .99. Meade and Islam (1995) have shown that the standard logistic curveweuse(alongwithanothersimplespecification) outperforms more complicated models. This is largely because allowing for non-linearity and/or asymmetry around the inflection point asks too much of the limited

19 Figure 3: Debit Card Transaction Shares for Norway and the Netherlands

1.0 NORDCARDPRED × YEARP Spline k=7.00 NLDCARDPRED × YEARP Spline k=7.00 0.9

0.8

0.7

0.6 Norway Debit Card Transactions Actual data: 1990-2004 0.5 Netherlands Debit Card Transactions 0.4 Actual data: 1990-2004

0.3

0.2

0.1

0.0

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

data points in the figures are predicted shares from a fitted logistic curve. An alternative way to judge the effect of pricing payment services is to compare the slopes of the estimated S-curves. For debit cards in Figure 3, b = .197 for Norway while b = .249 and is higher for the Netherlands, suggesting a more rapid adoption rate. While this difference is statistically significant at the .05 (but not .01) level, it is not very important economically (as seen visually from the figure and noted in Table 7 where the yearly average change in shares are shown). For electronic giro transactions in Figure 4, the difference in slopes is greater (b =.202 for Norway, b = .126 for the Netherlands) and this difference is significant at the .01 level, as can be inferred simply by looking at the graph.32 From a public policy perspective, this suggests that consumer valuation of the convenience, security, and other non-price benefits of debit card use is important and that pricing—given the relative prices being charged—has likely had only a relatively small incremental impact on the trade-off between debit card use versus obtaining cash from an ATM. In contrast, pricing appears to make a significant difference in the adoption rate of electronic giro transactions. This is consistent with the larger price elasticities usually estimated for electronic giros compared to debit cards. The discounted social benefit of electronic payments can be approximated by considering the difference in the bank cost of producing electronic versus paper payments (Row 1 of Table 8) and the speed by which a country shifts to electronic payments (determined, as an data typically available. 32Assuming the two S-curves are independent, since they refer to different countries, the means test for Figure 3was:t =(.249 .197)/ (.021)2 +(.006)2 =2.38 and was t =(.202 .126)/ (.010)2 +(.006)2 =6.52 for Figure 4. − − s s

20 Figure 4: Electronic Giro Transaction Shares for Norway and the Netherlands

1.0 NOREGIROPRED × YEARP Spline k=7.00 NLEGIROPRED × YEARP Spline k=7.00 0.9

0.8 Netherlands Electronic Giro Transactions 0.7 Actual data: 1990-2004

0.6

0.5 Norway Electronic Giro transactions Actual data: 1990-2004 0.4

0.3

0.2

0.1

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

approximation, from Figures 3 and 4).33 The per transaction bank cost figures used here are for Norway but are very similar to those for the Netherlands. As the time taken to shift from paper to electronics differed across countries and payment instruments, we computed benefits using a common number of years (40). We started when the share of electronics was .10 and ended when it reached .90, to eliminate the slow start and finish evident in the fitted S-curves which are farther from our observation set. Once a .90 share was attained, the undiscounted nominal benefits remained at that level until the end of the 40 year period. Benefits are computed using the level of payments in 1999, noted in the footnote to the table, as we do not have information on what that level was before 1990 nor after 2004. The annual per person benefits were discounted at 3% and sums of the per person discounted (and undiscounted) benefits are shown in Table 8. Shifting to debit cards from ATMs apparently saves €945 perpersoninNorwayand€769 in the Netherlands while the shift to electronic from paper giro payments saves €814 and €881 per person, respectively, in the two countries. Factoring in the savings from eliminating checks adds €461 and €676 in savings per person in Norway and the Netherlands, discounted over the same 40 year period.34 Overall, the apparent

33Savings in merchant costs are typically unknown but, if estimated, should be included (taking care not to double count certain expenses, such as including bank fees in merchant costs). Some merchant cost savings information does exist for point of sale transactions in the Netherlands but no information exists for merchant savings for giro payments (Brits and Winder, 2004). 34Bank check costs per transaction were 2.70 euro while debit card expenses were .30 euro. This gives a net savings of 2.40 euro for each of the 11.8 (Norway) and 17.3 (Netherlands) checks written per person in 1990 that were fully replaced by debit cards by 2004. Had the net savings per check been the same as the cost diﬀerence between debit cards and ATMs (1.01 euro - .30 euro = .71 euro), then the values shown in Table 8 would already include the check savings. As the check savings are larger, the additional savings from checks foreachyearis2.40euro-.71euro=1.69eurotimestheperpersonchecknumbersabovetimes40yearsand

21 Table 8: Estimated Per Person Beneﬁt from Electronic Payments (all values are in euros) Debit Card vs. ATM Electronic vs. Paper Giro Norway Netherlands Norway Netherlands Cost Diﬀerence 1.01 - .30 1.01 - .30 1.39 - .70 1.39 - .70

Undiscounted 1,907 1,533 1,644 1,835 Discounted 945 769 814 881 Per person debit card + ATM transactions in 1999 were 95 per year in Norway and 72 in the Netherlands while electronic + paper giro transactions were, respectively, 84 and 120. discounted savings in bank costs per person from electronic payments is €2,220 in Norway and €2,326 in the Netherlands. If the switch to electronic payments took place immediately rather than over a 40 year horizon, the total savings would be €0.7 billion for all of Norway (.35% of GDP in 2004) and €2.9 billion for all of the Netherlands (.61% of GDP).35

6 Summary and Conclusions.

Electronic payment instruments (credit cards excepted) are considerably cheaper than their paper-based alternatives, including cash. Banks and merchants are interested in shifting users to electronic payments to save costs, as are some government policy makers who seek to improve the cost efficiency of their nation’s payment system. Historically, banks have recouped their payment costs through: (1) interest earned on payment float (from delaying availability of funds credited to accounts and debiting accounts prior to bill payment value dates); (2) maintaining a spread between market rates and the rate paid on deposits; and (3) charging flat monthly fees or imposing balance requirements. In contrast to business users, consumers face very few payment services that are priced on a per transaction basis and so have little incentive to choose the lowest cost instrument either at the point of sale or for bill payments. Banks are well aware that transaction pricing can speed up the shift to electronic payments but are reluctant to lose deposit market share by being the first (and perhaps only) bank to implement explicit prices differentiated according to underlying costs. While this problem is mitigated if most (or all) banks implement pricing at about the same time, antitrust authorities are unlikely to view such coordination as being in the public interest unless the social benefits from pricing are significant and the quid pro quo is a compensating reduction in payment float, a higher interest rate paid on deposits, or a reduction in flat fees or balance requirements. Indeed, float reduction was the trade-off when banks coordinated the timing of when they would implement pricing in Norway (there was no coordination in the prices to be charged and initially some were zero). then discounted at 3%. This gives the discounted sums reported for checks in the text. 35Multiplying the estimated cost difference for debit cards and ATMs in Norway and the Netherlands (.71 euro) by the number of total debit card and ATM transactions per person in 1999 (95 and 72, respectively) and then by the population of each country (4.6 and 16.3 million, respectively) gives the cost savings estimate if debit cards immediately replaced ATMs. The cost difference for debit cards and checks is 2.40 euro and the respective number of transactions in 1990 was 11.8 and 17.3 checks per person. The cost difference of electronic versus paper giro payments was .69 euro while the respective number of transactions was 84 and 120 per person in 1999. These values, plus GDP, were used to compute the estimated savings in the text. Including merchant cost savings would raise these values.

22 In this paper we use the experience of Norway (which priced its payment services) and the Netherlands (which did not) over 1990-2004 to try to determine what the incremental effect of transaction pricing may be on the adoption of debit cards versus withdrawing cash from an ATM and on the adoption of electronic giro transactions (credit transfers and direct debits) over paper giros. Specifically, we compare payment instrument use per person in Norway in response to the prices being charged, the availability of terminals, and the level of real consumption with the experience of the Netherlands which also adopted electronic payments but did not price. Our four equation country difference model spanned 15 years— the limit of the available data—and during this time the share of electronic payments rose by some 60 percentage points, from around the mid-twenties to the mid-eighties which in most cases easily covered the inflection point where the share of electronic payments switches from rising at an increasing rate to rising at a decreasing rate. Our model is estimated in a systems equation framework using levels data and robustness is illustrated by estimating models in a first difference and error correction framework. Price and terminal elasticities derived from these models form the basis for our conclusions and indicate the incremental effect of pricing on the adoption rate of electronic payments. We also attempt to estimate the social benefit of shifting to electronic payments and offer guidance on how pricing may best be implemented. The effects of pricing differ depending on which instruments are being considered. Over- all, pricing has a smaller effect on shifting consumers from ATM cash withdrawals to debit card use than it does in shifting use from paper to electronic giro transactions. The reason for this difference seems to be that there are non-price benefits associated with debit card use (convenience, security) that consumers value such that the availability of terminals needed to function debit card transactions have a stronger effect on debit card use than prices, as evidenced by the fact that the debit card price elasticity is smaller than the terminal elasticity. Debit cards also substitute for costly checks and the high price on these instruments in Norway was associated with their virtual elimination, although the same thing happened in the Netherlands which did not price. While terminal availability appears to have a stronger effect on debit card use than does pricing, the shift to cards can be speeded up when pricing is combined with terminal availability. Using our estimated elasticities and the actual changes in prices and terminals, the predicted relative rise of debit card use over ATMs was 8.0% from terminal effects alone but rose to 10.4% with pricing, an increase of around 20%. The effect of pricing on electronic giro use was greater than it was for debit cards since the electronic giro price elasticity is larger and the percent change is price experienced was greater. Reasons for this difference are the above-mentioned non-price convenience and security attributes of debit cards along with the fact that for one-third of our time period the absolute price of debit cards was higher than the weighted average price of an ATM cash withdrawal. In contrast, the price of an electronic giro was always absolutely lower than the paper giro price. Even though the relative price of debit cards and electronic giros were both falling over the entire period, the higher absolute price of a debit card transaction versus an ATM would be expected to dull the overall price response being measured for the entire period since there is no strong reason to believe that the price response is symmetric (and symmetry was not imposed in our model) since the non-price attributes of debit cards and ATMs are different. Thus if pricing is implemented, it will likely be more successful if the absolute price of the less expensive instrument is always absolutely lower per transaction than the price of the more expensive instrument.36

36This was not done in Norway, perhaps because dispensing cash via an ATM was already less expensive

23 In terms of cost savings, the shift from ATM cash withdrawals and checks to debit cards plus the savings from shifting from paper to electronic giro transactions —if it happened without a lag— may save €0.7 billion in bank cost for Norway (.35% of GDP in 2004) and €2.9 billion for the Netherlands (.61% of GDP).37 Merchant cost savings, for which little information exists, would increase these savings estimates as would viewing the discounted bank cost savings over a 40 year transition period (and is noted in the text). As both of these countries are well on their way to realizing the full potential gains from electronic payments, the issue of pricing or not pricing is a policy topic for developed countries that are not as far along in the substitution process or for most developing countries that are just in the initial stages of thinking about how to improve the eﬃciency of their payments system. The social beneﬁts of electronic payments are quite large and may convince antitrust authorities to allow the coordination of the timing of the implementation of pricing (but not of course the prices to be charged) to speed up this transition if banks wished to adopt explicit pricing. And with interest rate margins falling, worsening the recoupment of bank payment costs, this wish may soon become a reality.

than dispensing it through a branch oﬃce (assuming the rise in dispensing frequency at an ATM does not rise enough to oﬀset this advantage). 37The savings are absolutely higher for the Netherlands primarily because their population (16.3 million) is much larger than that for Norway (4.6 million), but GDP per capita is lower.

24 7 Bibliography.

Bolt, W., and A.F. Tieman, “Skewed Pricing in Two-Sided Markets: An IO approach”, DNB Working Paper 2004/13, De Nederlandsche Bank.

Brits, H., and C. Winder, “Payments Are No Free Lunch”, Mimeo, De Nederlandsche Bank, 2005,.

De Nederlandsche Bank, “The Cost of Payments”, Quarterly Bulletin, March 2004: 57-64.

Flatraaker, D.-I., and P. Robinson, “Income, Costs, and Pricing in the Payment System”, Norges Bank Economic Bulletin, 66, 1995: 321-332.

Gresvik, O., and G. Øwre, “Costs and Income in the Norwegian Payments System 2001. An Application of the Activity Based Costing Framework”, Working Paper 2003/8, Norges Bank.

Humphrey, D., M. Kim, and B. Vale, “Realizing the Gains from Electronic Payments: Costs, Pricing and Payment Choice”, Journal of Money, Credit and Banking, May 2001, 216-234.

Humphrey, D., M. Willesson, G. Bergendahl, and T. Lindblom, “Beneﬁts from a Changing Payment Technology in European Banking”, Journal of Banking and Finance,forthcoming 2005.

Norges Bank, “Annual Report on Payment Systems”, 2004, 2000, 1997, Norway.

Meade, N., and T. Islam, “Forecasting with Growth Curves: An Empirical Comparison”, International Journal of Forecasting, 11, 1995: 199-215.

Rochet, J-C., and J. Tirole, “Platform Competition in Two-Sided Markets”, Journal of the European Economic Association, 1, 2003: 990-1029.

Sveriges Riksbank, “Prices and Costs in the Swedish Payment System”, in Financial Stability Report, 2004: 65-73.

Zinman, J., “Why Use Debit Instead of Credit? Consumer Choice in a Trillion-Dollar Market”, Federal Reserve Bank of New York Staﬀ Reports, no. 191, July 2004.

25 8 Data Appendix.

Parameter estimates and t-statistics are shown below for the system estimation of equations (1) to (4). Standard Errors were computed from a heteroscedastic-consistent matrix (Robust- White).38

Variable Parameter Estimation T-statistic

Constant α1 .497961 2.77398 DIFDCARDTML α2 .373649 2.55827 DIFDCARDPRICE α3 .055702 .180209 DIFDCARDTML2 α22 .167924 1.24453 DIFDCARDPRICE2 α33 .870318 1.21951 DCARDTMLPRICE α23 -.308628 -.922949 DIFPCONS α4 -.571509 -1.13726 Constant β1 .035974 .554554 DIFATMTML β2 -1.67562 -8.56053 DIFDCARDPRICE β3 1.00388 5.85276 DIFATMTML2 β22 5.93594 7.78129 DIFDCARDPRICE2 β33 2.33231 3.40630 ATMTMLPRICE β23 -3.15162 -4.50450 DIFPCONS β4 -.138567 -.766170 Constant γ1 -.519080 -1.21248 DIFEGIROPRICE γ2 1.06886 2.15115 DIFEGIROPRICE2 γ22 .881313 3.06231 DIFPCONS γ3 .544589 1.13520 Constant δ1 -.606181 -1.20153 DIFEGIROPRICE δ2 -1.69478 -2.90182 DIFEGIROPRICE2 δ22 -1.13584 -3.30711 DIFPCONS δ3 -.678980 -1.34759

Debit Card equation (1): R2 = .978, Durbin-Watson = 2.12 ATM Cash Withdrawal equation (2): R2 = .987, Durbin-Watson = 2.80 Electronic Giro equation (3): R2 = .872, Durbin-Watson = 1.86 Paper Giro equation (4): R2 = .802, Durbin-Watson = 2.03

Data description and sources:

The Netherlands: Source: DNB Statistics (www.dnb.nl), CBS Statistics (www.cbs.nl) and National Accounts statistics.

38Our data computations follow a translog type of approach where squared values of variables are computed as (ln X)2 rather than ln (X2) and interaction variables are (ln X)(ln Y) rather than ln (XY). Consequently, 2 2 2 2 ﬁrst diﬀerences are computed as (ln X)t (ln X)t 1 rather than ln (X )t ln (X )t 1 and interaction variables − − − − are (ln X)(ln Y)t (ln X)(ln Y)t 1 rather than ln (XY)t ln (XY)t 1. The two approaches give similar but − − not identical results.− −

26 Debit card transactions include goods purchased with domestic debit cards, credit card usage (of about 3 percent of total card usage) is not included. Debit card terminals are EFTPOS terminals at the point of sale in retail locations. ATM cash withdrawal transactions only include withdrawals with domestic debit cards. ATM terminals are owned by Dutch comer- cial banks. Paper-based giro transactions include ordinary credit transfers (“mail giro”) and inpayment transfers (“accept-giro”). Electronic giro transactions include credit transfers, standing order credit transfers and direct debits. Consumption is derived from National Accounts data on ﬁnal private consumption expenditures, per capita, and corrected for inﬂation using CPI data. Demographic data are constructed using data from the Dutch Central Bureau of Statistics.

Norway: Source: Norges Bank (www.norgesbank.no), Norges Bank Annual Report on Payment Sys- tems (2004, 2000, 1997), Statistics Norway (www.ssb.no), and International Finance Statistics database.

Debit card transactions include goods purchased with Norwegian payment cards in Norway and abroad, but does not include use of oil company cards (see footnote 4 in the text). Debit card terminals are EFTPOS terminals that also include terminals owned by oil companies (since all debit cards can be used). ATM cash withdrawal transactions include withdrawals from commercial banks and savings banks. These banks own the ATM terminals. Debit card price relates to the bank per transaction price for payment cards using an EFTPOS terminal. ATM withdrawal price is a weighted price for a cash withdrawal, based on observed prices and volume weights corresponding to withdrawals at bank’s own ATM terminals during opening hours and outside of opening hours, and other bank’s ATM terminals during opening hours and outside of opening hours. Paper-based giro transactions include mail giro, giro for cash payments, account debits, and money orders. Electronic giro transactions include company terminal giros, giro via internet or telephone, credit transfers, and direct debits. Paper-based giro price is a weighted price per transaction using observed prices and volume weights corresponding to mail giro, giro for cash payments, account debits, and money orders. Electronic giro price is a weighted price per transaction using observed prices and volume weights for terminal giro, internet giro, telephone giro, and direct debits. Consumption is derived from IFS data on ﬁnal private consumption expenditures, per capita, and corrected for inﬂation using CPI data. All nominal values and prices are converted into euros using a purchasing power parity exchange rate published by Norges Bank. Demographic data are constructed using data from Statistics Norway.

Some missing values at the beginning of the sample were estimated by applying simple ex- trapolation procedures.